Cognitive Interference Mitigation Power Control

Power control schemes can either be decentralised or centralised as presented in the following subsections.

2.2.1.1 Decentralised Power Control

A scheme which requires the exploitation of a MBS control channel by an FAP and FUEs in a network to mitigate interference is presented in [79]. By using CR, a FAP and its UEs are able to decode control information such as number, location and power of each active MUE in a spectrum in order to adjust their transmit power. Two rules are implemented after the

Cognitive Interference Mitigation Power Control Centralised Power Control Decentralised Power Control Spectrum Access Frequency based Time based Joint Frequency and Time based

Centralised / Decentralised Schemes Individual vs Group Channel Sensing Antenna Schemes Single Element vs Multi Element Adaptive beam forming vs Adaptive Pattern Switching Joint Schemes

acquisition of the necessary control information to mitigate interference. As a first rule, the MBS and MUEs are given top transmission priorities in what is known as forbidden radius which is the area of an active MUE or a MBS. FAPs within the range of this forbidden radius are restricted to transmit. In the second rule, FAPs and FUEs are tasked with controlling their transmit power such that the interference received at the MBS or MUE does not exceed a set value. A MUE assisted power control scheme is adopted in [80] where MUE measures the received power from its serving MBS and forwards the information to all FAPs in its vicinity. Each FAP subsequently calculates its path loss from the MUE utilising CR to optimize its power level to avoid interference with the MUE.

2.2.1.2 Centralised Power Control

As mentioned earlier, FAPs can be deployed in an open, closed and hybrid access which largely depend on the subscriber’s choice. A novel approach with the ability of FAPs to switch between these access modes based on cognitive sensing and power control is presented in [81]. This self-configurability approach requires each FAP to sense the radio environment and identify white spaces or slots with less interference. For each slot identified, a SINR threshold is set based on channel conditions and a power control algorithm updates the transmit power to define its coverage range.

This predefined SINR threshold determines the change in access mode. For example, if a received SINR for a particular slot is less than this threshold it means it can accommodate unregistered UEs in its vicinity and therefore switches to open/hybrid access as long as it does not affect the FUEs being served by the FAP otherwise it switches to a closed/hybrid access to limit the UEs. This scheme highlights the dynamic capabilities of the femtocell. However, in reality, most subscribers would prefer a single CSG mode as added security becomes a factor in open or hybrid modes. Moreover, femtocells are paid for by the subscribers who would not want to share resources with unknown users.

A power control scheme using Q-learning which enables FAPs to allocate power optimally in a cognitive underlay approach is proposed in [82] to mitigate cross-tier interference in the downlink. A FAP carries out a distributed learning technique by sensing the radio environment to observe its state and takes an action to determine its consequences which can be assessed as a reward (low interference and high MUE capacity) or penalty (high interference and low MUE capacity). By repeating this process, it analyses the entire radio environment and is able to find an optimal power allocation policy to mitigate interference while maintaining MUE capacity. The problem with this scheme is that by trying to determine a suitable policy to mitigate interference, it accumulates a lot of signalling overhead which also leads to delay.

Another power control scheme utilizing communication in the uplink (UL) and composed of three phases (channel sensing, channel training and data transmission) is described in [83]. During channel/spectrum sensing, a FAP senses the radio environment by employing any of the well known techniques to find unoccupied spectrum. A hypothesis test is conducted to make a decision as to whether a spectrum is occupied (MUE(s) present) or Null (absent). The channel coefficient between the FAP and the FUE is estimated in channel training by the FUE sending a signal known as training signal to its FAP. The cognitive FAP thus optimises the rate at which power is transmitted by allowing its FUE transmit at a reduced or maximum power when a MUE is present or absent respectively to avoid interference during data transmission.

The technique proposed in [84] is based on the overlay approach where each FAP in the network periodically senses and deduces the macrocell path loss denoted as PLM, between itself and the MBS. Spectrum sensing in this case is divided into two stages, the uplink and downlink sensing. In the uplink stage, FAP deduces the PLM of MUE by measuring a parameter such as Reference Signal Received Power (RSRP). In downlink sensing, based on the PLM, the spectrum sensing threshold denoted as γ is calculated. Using this information, a FAP is able to identify a channel as unoccupied if the received signal power, denoted as PMC,

on that channel does not exceed γ. On each sensed channel, FAP is able to allocate a lagrangian based transmit power function on its FUE for data transmission to mitigate interference.

In a nutshell, most power control schemes utilize an adaptive power control mode where the pilot power levels of an FAP is controlled effectively not only to mitigate interference but to reduce the need for handover of close by UEs in open access mode. Although UEs can also employ power control techniques, distributed power control schemes assisted by UEs of a FAP or MBS may have a detrimental effect on the UEs as it increases their overhead which may result to increased battery drain. On the other hand, a centralised power control scheme will require a continuous update of the information about all its UEs in real time which makes it computationally complex. In our opinion, since most UEs are mobile devices it is better to leave the optimisation of power to FAPs and MBS (centralised) who have a dedicated power supply and make the computation of the algorithms less complex.